(2.2)
eans if
O Store
ps of 5
ecially
needed
els the
hus we
ber of
sitivity
ly one
a new
1ce. At
shown
image.
rk and
> easily
in the
> With
2.2 the
of the
] result
mosaic
of both
Stefan Diener
After the orientation of all separate colour channels to the mosaic image a single RGB image can be created. The low
resolution RGB image and the high resolution mosaic image are covering the same area on the ground.
To prepare the colour image for the pixel-wise colour space transformation the image has to be scaled up to the size of
the mosaic image. We investigated the bilinear and bicubic interpolation methods.
To make an objective comparison we scaled down an image (to 1/4 of width and height), scale up the intermediate
image to the original size and calculated the difference between the start image and the down-up-scaled image. This was
done with a couple of images. We compared the resulting unbiased RMSE and the required time for scaling up for both
bilinear and bicubic interpolation.
The RMSE of the bicubic interpolated images is 2-4% less, but the algorithm is 15-30% slower than the bilinear
interpolation. Remembering the significant better visual results for a few sample images the smaller RMSE becomes
more important than the 2-4% ratio implies. We will use the bicubic and bilinear interpolation method for a high quality
image processing (i.e. for scaling and affine transformations).
| red [ green j blue HighRes
| channel channel channel mosaic image
* 7 EN pr. *.
affine
transformations
B M AV
transformed | { transformed | transformed
i red channel green channel blue channel
i J
join into à. ^« f
single image —
scale up
{interpolated "HighRes" |
| RGB image
s olour spac EN.
^ transformation ^
- p 4 i A S
| pure luminance Í pure colour
L information | information
N { 2
S inverse colaur
Ro space
x transformation /
x
^
| HighRes RGB
mosaic image
Figure 2. Principal workflow of the colour composite generation.
After scaling up the RGB image to the size of the mosaic image one has to separate the luminance and the pure colour
information. This is done by transforming the image from the RGB colour space into another suitable colour space as
described in the following section.
Now the fusion can be performed: The extracted luminance from the interpolated RGB image has to be replaced by the
mosaic image. To complete the workflow the inverse colour space transformation is necessary. The resulting image is
the desired high resolution RGB mosaic image, which is ready for further processing steps.
32 Comparison of colour space transformations
We compared the following colour spaces and their belonging transformations from and to RGB:
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part Bl. Amsterdam 2000. 85